Building GHDL from Sources#

GHDL can be downloaded as a tarball/zipfile or cloned with git clone from GitHub. GitHub offers HTTPS and SSH as transfer protocols. See the Sources page for further details.

Important

Since GHDL is written in Ada, independently of the code generator you use, a compiler is required. Most GNU/Linux package managers provide gcc-ada or gcc-gnat (which could be outdated). Alternatively, GNU Ada compiler, GNAT GPL, can be downloaded without registration from libre.adacore.com (2014, or later; for x86, 32 or 64 bits).

Hint

The download page of GNAT Community Edition provides the latest version (x86, 64 bits), with a graphical installer (chmod +x *.bin and execute it). Alternatively, you can find a link to More packages, platforms, versions and sources at the bottom of the page, where versions previous to 2018 are available as binaries ready to be installed (untar and run the doinstall script). In any case, you must add <GNAT_INSTALL_DIR>/bin to your PATH.

Attention

Since v0.37, GHDL’s synthesis features require GCC >=8.1, due to some new GNAT features which are not available in previous releases. Users with older versions (who don’t need synthesis) can configure GHDL with option --disable-synth.

GHDL currently supports three different back-ends (code generators):

  • mcode - built-in in-memory x86 (or x86_64) code generator

  • GCC - Gnu Compiler Collection (gcc.gnu.org)

  • LLVM - Low-Level Virtual Machine (llvm.org)

Here is a short comparison, so that you can choose the one you want to use:

Back-end

Pros

Cons

mcode

  • Very easy to build

  • Very quick analysis

  • Can handle very large designs

  • Base simulation time can be modified for speeding up execution

  • Simulation is slower

  • x86_64/i386 only

LLVM

  • Generated code is faster

  • Generated code can be debugged (with -g)

  • Easier to build than GCC

  • Ported to many platforms (x86, x86_64, armv7/aarch64)

  • Build is more complex than mcode

GCC

  • Generated code is faster (particularly with -O or -O2)

  • Generated code can be debugged (with -g)

  • Ported to many platforms (x86, x86_64, PowerPC, SPARC)

  • Build is even more complex

  • Analysis can take time (particularly for large units)

  • Code coverage collection (gcov) is unique to GCC

Hint

The output of both GCC and LLVM is an executable file, but mcode does not generate any. Therefore, if using GCC/LLVM, the call with argument -r can be replaced with direct execution of the binary. See section Simulation.

After making your choice, you can jump to the corresponding section. However, we suggest you to read Directory structure first, so that you know where the content will be placed and which files are expected to be created.

Hint

In these instructions, the configure script is executed in the source directory; but you can execute in a different directory too, like this:

$ mkdir ghdl-objs
$ cd ghdl-objs
$ ../path/to/ghdl/configure ...

Hint

On Windows, building GHDL with mcode backend and GNAT GPL 32 bit seems to be the only way to get a standalone native executable straightaway. MINGW/MSYS2 builds depend on the environment/runtime. See #1560.

Hint

For MacOS 10.15 (Catalina), see #1368 for workarounds to link failures.

TL;DR#

In order to follow the traditional way to configure and make, you need an Ada compiler.

Hint

Depending on the OS and distribution you are using, you will also need to install some toolchain dependencies, such as zlib.

To use mcode backend (easiest to build), in the GHDL base directory, configure and build:

$ ./configure --prefix=/usr/local
$ make

At that place, you can already use the ghdl_mcode built in the directory. You can also install GHDL:

$ make install

That’s all!

Hint

The executable is installed as ‘ghdl’ in /usr/local. To install it to a different path, change the --prefix in the call to configure. For example, on Windows, you may want to set it to --prefix=/c/Program Files (x86)/GHDL.